Reproduce “Signed but Not Safe: Requirement Integrity in Multi-Agent LLM Handoffs” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Xiaobo Chen, Genliang Zhu, Ziyuan Wang, Chu Wang
The authors introduce signed manifests and code-based checks to carry mandatory requirements through multi-agent workflows. Each requirement retains its verbatim wording and a checkable rule; signatures detect in-transit changes, while code checks the decision. On Qwen3-8B, unsafe commits under three attack conditions fell from 69.4% with structured JSON to 9.0% with signed manifests, but pre-signing corruption remains outside this protection.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.